Prof. Mireille Boutin
 

Machine Learning

The basis of this work is a 1D random projection approach to solve supervised/unsupervised pattern recognition problems. This approach is particularly well-suited for dealing with small datasets in high-dimension (fat data). The fact that this approach works is due to certain geometric properties of real, high-dimensional data. We build on this to propose benchmarks to quantify the complexity of a given pattern recognition problem. We also explore and construct different mathematical  models that shed light on these geometric properties.


  1. • M. Boutin, A. Bradford, “A highly likely clusterable dataset with no cluster,” arXiv preprint arXiv:1909.06511 [stat.ML]


  1. •T. Yellamraju, M. Boutin, “Clusterability and Clustering of Images and Other “Real” High-Dimensional Data,” IEEE Transactions on Image Processing, Vol. 27, No. 4, April 2018, pp. 1927 - 1938.


  1. •K. Larson, M. Boutin, “Performance Benchmarks for Detection Problems,” IEEE Global Conference on Signal and Information Processing (GlobalSIP), Montreal, QC, Canada, November 14-16, 2017.


  1. •S. Han and M. Boutin, “The Hidden Structure of Image Datasets,” IEEE International Conference on Image Processing (ICIP), Quebec City, Canada, September 27–30, 2015.


  1. •T. Yellamraju, J. Hepp, M. Boutin, “Benchmarks for Image Classification and Other High-dimensional Pattern Recognition Problems,” arXiv manuscript #1806.05272, 2018.


  1. •T. Yellamraju, M. Boutin, “Pattern Dependence Detection using n-TARP Clustering,” arXiv manuscript #1806.05297, 2018.


  1. •K. Larson, M. Boutin, “TARP Benchmarks for Detection Problems.”